Vertical Discretization for a Nonhydrostatic Atmospheric Model Based on High-Order Spectral Elements

• Published in: Monthly weather review Vol. 148; no. 1; pp. 415 - 436
• Main Authors: Yi, Tae-Hyeong, Giraldo, Francis X.
• Format: Journal Article
• Language: English
• Published: Washington American Meteorological Society 01.01.2020
• Subjects: Accuracy; Atmospheric models; B spline functions; Computer applications; Convergence; Coordinate transformations; Discretization; Mathematical models; Methods; Mountains; Orography; Polynomials; Spectra
• ISSN: 0027-0644; 1520-0493
• DOI: 10.1175/MWR-D-18-0283.1

Abstract This study addresses the treatment of vertical discretization for a high-order, spectral element model of a nonhydrostatic atmosphere in which the governing equations of the model are separated into horizontal and vertical components by introducing a coordinate transformation, so that one can use different orders and types of approximations in both directions. The vertical terms of the decoupled governing equations are discretized using finite elements based on either Lagrange or basis-spline polynomial functions in the sigma coordinate, while maintaining the high-order spectral elements for the discretization of the horizontal terms. This leads to the fact that the high-order model of spectral elements with a nonuniform grid, interpolated within an element, can be easily accommodated with existing physical parameterizations. Idealized tests are performed to compare the accuracy and efficiency of the vertical discretization methods, in addition to the central finite differences, with those of the standard high-order spectral element approach. Our results show, through all the test cases, that the finite element with the cubic basis-spline function is more accurate than the other vertical discretization methods at moderate computational cost. Furthermore, grid dependency studies in the tests with and without orography indicate that the convergence rate of the vertical discretization methods is lower than the expected level of discretization accuracy, especially in the Schär mountain test, which yields approximately first-order convergence.